YOLOv8-lite: An Interpretable Lightweight Object Detector for Real-Time UAV Detection

Hawking Lai, Bowie Liu, Ho Yin Kan, Chan Tong Lam, Sio Kei Im

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

UAV detection is an important problem in sensitive areas involving security and privacy. This paper proposes an interpretable lightweight model designed explicitly for the real-time detection of UAVs, called YOLOv8-lite. By employing a high-speed YOLOv8 model and Depthwise convolution, the model performs better than the original YOLOv8 with fewer parameters in the Det-fly dataset. The proposed YOLOv8-lite achieves impressive results with 0.98 AP50 and 0.68 AP0.5:0.95 on the test set, using only 2 million parameters. Meanwhile, YOLOv8-lite shows good results in solving the challenges of detecting UAVs against various environmental backgrounds. In addition, interpretability methods are applied to illustrate the factors contributing to the effectiveness and generalization capability of the model. The code for the model is available: https://github.com/hawkinglai/uav-det.

Original languageEnglish
Title of host publication2023 9th International Conference on Computer and Communications, ICCC 2023
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1707-1713
Number of pages7
ISBN (Electronic)9798350317251
DOIs
Publication statusPublished - 2023
Event9th International Conference on Computer and Communications, ICCC 2023 - Hybrid, Chengdu, China
Duration: 8 Dec 202311 Dec 2023

Publication series

Name2023 9th International Conference on Computer and Communications, ICCC 2023

Conference

Conference9th International Conference on Computer and Communications, ICCC 2023
Country/TerritoryChina
CityHybrid, Chengdu
Period8/12/2311/12/23

Keywords

  • Depthwise convolution
  • Interpretable machine learning
  • Object detection
  • UAV detection
  • YOLOv8

Fingerprint

Dive into the research topics of 'YOLOv8-lite: An Interpretable Lightweight Object Detector for Real-Time UAV Detection'. Together they form a unique fingerprint.

Cite this